This may betray my ignorance, but it seems kind of silly to try to create artificial intelligence that has no ability to sense/interact with its environment and boundaries. If we want true non-biological intelligence shouldn't we start with something simple that can "sense" its environment, and try to create something in machine code or assembly?
If we want to replicate life, shouldn't we replicate the system of small, mostly self-contained, life-like systems that form a part of a larger life-like system?
